Alfa Kilo Papa

[alfakilopapa]

Running Vizag Full Marathon

On Nov 12, 2017, I ran my first full marathon at Vizag navy marathon. This blog is a summary of my findings from my run and post-hoc analysis of the route. Vizag city has unusual elevation profile due to eastern hills and coast line. The marathon course has several uphill and downhill sections. The elevation profile is illustrated below. To contrast, I also added New York City Marathon elevation profile.  One can clearly see the elevation gains are sharp in Vizag course as opposed to New York one. Thus, Vizag navy marathon is one of the challenging courses for marathon runners. 

Disclaimer: I did not run NYC marathon and I only use its route profile as a benchmark to put Vizag marathon course in the context. 

Vizag marathon course is a roundtrip between Vishwapriya Function Hall on beach road and INS Kalinga (eastern naval command headquarters). There were 4 categories in the race -- 5k, 10k, Half-M, Full-M. According to organizers there are 15,000 participants in this year’s edition.  I guesstimate 250 participants ran in Full-Marathon category.  The race began at 4:30 am and I managed to reach half-way in 2 hrs 10 mins and finish line in 5 hrs 40 mins. A huge shoutout to the volunteers who were stationed for every 2 km. They were stocked up with bananas, water and glucose! Thanks to organizers.

courtesy: Vizag Navy Marathon website

My Training

Last time when I trained for half-marathon in Brooklyn, I regularly ran for 50 km a week on average. This time around, I primarily spent my time on strength-training and core-workouts  over several weeks before marathon. To be honest, my weekly mileage was adequate (about 50-60 km) for a half-marathon preparation but not a full marathon. I did a lot of deadlifts, squats, lunges, speed-biking, lateral running, zero-runner, elliptical and other large-muscle group routines. Hindsight, I should have run more long runs. I promise myself that I will focus on mileage for my next full marathon.

What was going through my mind

I was mostly nostalgic (because its my hometown), thoughtless, in-the moment throughout the race. It was a 4:30 am start and days into the race day, I regularly woke up at 3 am. I was excited to run! On the race day, I woke up at 2 am and had curd rice for breakfast. Once the race began, I started off slow and easy for the first 5 km. Once my cardio-vascular system warmed up, I picked up the speed and ran comfortably at ~9 min/mi for the next 10 km. I thoroughly enjoyed the eastern sea breeze while running north. It was still dark and I could barely see twilight for the first 12 km. I saw the sun rise over Bay of Bengal at 18th km. It was amazing! I wasn’t running for the most part of 21k, I was just sailing :) 

The course takes a U-turn at 21st km and we run back to the start point. The sun was out but not quite bright at this time. I started to pace myself around ~11 min/mi for the next 6 km (27th km). I managed to run the 21K at comfortable pace and reached halfway by 2 hrs 10 minutes. Until 27th km and the sun was not up it was more or less comfortable.   I started to feel the cramps in my calves once the uphill sections began. I mostly walked through the uphill sections and ran whenever I could. I smiled at everyone and anyone while I was walk/run-ing to keep my spirits up. I felt like a zombie with nothing on mind except watching the road and random strangers.  Despite being a flat section, the last 3 km was most grueling and the finish line was so near yet so far. I was greeted by my grandfather  at the finish line.

Elevation Profile for Vizag Marathon 

In Vizag marathon, there’s only 8.82 km of flat section in the entire route. The longest hill section  is 1.08 km, which is around Rushikonda and max height gain is 32 m. Based on the route profile below, one can observe that there are several hill sections across the route. In contrast, only few hill sections appear in NYC race course.

NYC elevation profile provided below for contrast

courtesy: plotaroute.com

Route Statistics for Vizag Marathon

route statistics for NYC  Marathon for contrast

NYC marathon begins in Staten Island and ends in Midtown. In contrast to Vizag course, NYC marathon course has a flat section of 9.90 km and the longest hill section of 0.81 km appears at the crossing between Brooklyn and Queens. Although total uphill sections of both courses are similar 16.56km vs 16.74km, the uphill sections are almost evenly distributed in Vizag, which can easily throw your pace off. 

[yahoo]

Important Security Information for Yahoo Users

By Bob Lord, CISO

Following a recent investigation, we’ve identified data security issues concerning certain Yahoo user accounts. We’ve taken steps to secure those user accounts and we’re working closely with law enforcement.

What happened?

As we previously disclosed in November, law enforcement provided us with data files that a third party claimed was Yahoo user data. We analyzed this data with the assistance of outside forensic experts and found that it appears to be Yahoo user data. Based on further analysis of this data by the forensic experts, we believe an unauthorized third party, in August 2013, stole data associated with more than one billion user accounts. We have not been able to identify the intrusion associated with this theft. We believe this incident is likely distinct from the incident we disclosed on September 22, 2016.

For potentially affected accounts, the stolen user account information may have included names, email addresses, telephone numbers, dates of birth, hashed passwords (using MD5) and, in some cases, encrypted or unencrypted security questions and answers. The investigation indicates that the stolen information did not include passwords in clear text, payment card data, or bank account information. Payment card data and bank account information are not stored in the system the company believes was affected.

Separately, we previously disclosed that our outside forensic experts were investigating the creation of forged cookies that could allow an intruder to access users’ accounts without a password. Based on the ongoing investigation, we believe an unauthorized third party accessed our proprietary code to learn how to forge cookies. The outside forensic experts have identified user accounts for which they believe forged cookies were taken or used. We are notifying the affected account holders, and have invalidated the forged cookies. We have connected some of this activity to the same state-sponsored actor believed to be responsible for the data theft the company disclosed on September 22, 2016.

What are we doing to protect our users?

We are notifying potentially affected users and have taken steps to secure their accounts, including requiring users to change their passwords. We have also invalidated unencrypted security questions and answers so that they cannot be used to access an account. With respect to the cookie forging activity, we invalidated the forged cookies and hardened our systems to secure them against similar attacks. We continuously enhance our safeguards and systems that detect and prevent unauthorized access to user accounts.

What can users do to protect their account?

We encourage our users to visit our Safety Center page for recommendations on how to stay secure online. Some important recommendations we’re re-emphasizing today include the following:

Change your passwords and security questions and answers for any other accounts on which you used the same or similar information used for your Yahoo account;

Review all of your accounts for suspicious activity;

Be cautious of any unsolicited communications that ask for your personal information or refer you to a web page asking for personal information;

Avoid clicking on links or downloading attachments from suspicious emails; and

Consider using Yahoo Account Key, a simple authentication tool that eliminates the need to use a password on Yahoo altogether.

For more information about these security matters and our security resources, please visit the Yahoo Security Issue FAQs page, https://yahoo.com/security-update.

Statements in this press release regarding the findings of Yahoo’s ongoing investigations involve potential risks and uncertainties. The final conclusions of the investigations may differ from the findings to date due to various factors including, but not limited to, the discovery of new or additional information and other developments that may arise during the course of the investigation. More information about potential risks and uncertainties of security breaches that could affect the Company’s business and financial results is included under the caption “Risk Factors” in the Company’s Quarterly Report on Form 10-Q for the quarter ended September 30, 2016, which is on file with the SEC and available on the SEC’s website at www.sec.gov.

[yahooresearch]

Presenting an Open Source Toolkit for Lightweight Multilingual Entity Linking

By Aasish Pappu, Roi Blanco, and Amanda Stent

What’s the first thing you want to know about any kind of text document (like a Yahoo News or Yahoo Sports article)? What it’s about, of course! That means you want to know something about the people, organizations, and locations that are mentioned in the document. Systems that automatically surface this information are called named entity recognition and linking systems. These are one of the most useful components in text analytics as they are required for a wide variety of applications including search, recommender systems, question answering, and sentiment analysis.

Named entity recognition and linking systems use statistical models trained over large amounts of labeled text data. A major challenge is to be able to accurately detect entities, in new languages, at scale, with limited labeled data available, and while consuming a limited amount of resources (memory and processing power).

After researching and implementing solutions to enhance our own personalization technology, we are pleased to offer the open source community Fast Entity Linker, our unsupervised, accurate, and extensible multilingual named entity recognition and linking system, along with datapacks for English, Spanish, and Chinese.

For broad usability, our system links text entity mentions to Wikipedia. For example, in the sentence Yahoo is a company headquartered in Sunnyvale, CA with Marissa Mayer as CEO, our system would identify the following entities:

Yahoo – linked to https://en.wikipedia.org/wiki/Yahoo!

Sunnyvale, CA – linked to https://en.wikipedia.org/wiki/Sunnyvale,_California

Marissa Mayer – linked to https://en.wikipedia.org/wiki/Marissa_Mayer

On the algorithmic side, we use entity embeddings, click-log data, and efficient clustering methods to achieve high precision. The system achieves a low memory footprint and fast execution times by using compressed data structures and aggressive hashing functions.

Entity embeddings are vector-based representations that capture how entities are referred to in context. We train entity embeddings using Wikipedia articles, and use hyperlinked terms in the articles to create canonical entities. The context of an entity and the context of a token are modeled using the neural network architecture in the figure below, where entity vectors are trained to predict not only their surrounding entities but also the global context of word sequences contained within them. In this way, one layer models entity context, and the other layer models token context. We connect these two layers using the same technique that (Quoc and Mikolov ‘14) used to train paragraph vectors.

Architecture for training word embeddings and entity embeddings simultaneously. Ent represents entities and W represents their context words.

Search click-log data gives very useful signals to disambiguate partial or ambiguous entity mentions. For example, if searchers for “Fox” tend to click on “Fox News” rather than “20th Century Fox,” we can use this data in order to identify “Fox” in a document. To disambiguate entity mentions and ensure a document has a consistent set of entities, our system supports three entity disambiguation algorithms:

Forward Backward Algorithm (Austin et al. 91)*

Exemplar Clustering (Frey and Dueck ‘07)

Label Propagation (Talukdar and Crammer ‘09)

*Currently, only the Forward Backward Algorithm is available in our open source release–the other two will be made available soon!

These algorithms are particularly helpful in accurately linking entities when a popular candidate is NOT the correct candidate for an entity mention. In the example below, these algorithms leverage the surrounding context to accurately link Manchester City, Swansea City, Liverpool, Chelsea, and Arsenal to their respective football clubs.

Ambiguous mentions that could refer to multiple entities are highlighted in red. For example, Chelsea could refer to Chelsea Football team or Chelsea neighborhood in New York or London. Unambiguous named entities are highlighted in green.

Examples of candidate retrieval process in Entity Linking for both ambiguous and unambiguous examples referred in the example above. The correct candidate is highlighted in green. At this time, Fast Entity Linker is one of only three freely-available multilingual named entity recognition and linking systems (others are DBpedia Spotlight and Babelfy). In addition to a stand-alone entity linker, the software includes tools for creating and compressing word/entity embeddings and datapacks for different languages from Wikipedia data. As an example, the datapack containing information from all of English Wikipedia is only ~2GB.

The technical contributions of this system are described in two scientific papers:

Roi Blanco, Giuseppe Ottaviano, and Edgar Meij: “Fast and space-efficient entity linking in queries.” In Proceedings WDSM 2015.

Aasish Pappu, Roi Blanco, Yashar Mehdad, Amanda Stent, and Kapil Thadani: “Lightweight multilingual entity extraction and linking.” In Proceedings WSDM 2017.

There are numerous possible applications of the open-source toolkit. One of them is attributing sentiment to entities detected in the text, as opposed to the entire text itself. For example, consider the following actual review of the movie “Inferno” from a user on MetaCritic (revised for clarity): “While the great performance of Tom Hanks (wiki_Tom_Hanks) and company make for a mysterious and vivid movie, the plot is difficult to comprehend. Although the movie was a clever and fun ride, I expected more from Columbia (wiki_Columbia_Pictures).”  Though the review on balance is neutral, it conveys a positive sentiment about wiki_Tom_Hanks and a negative sentiment about wiki_Columbia_Pictures.

Many existing sentiment analysis tools collate the sentiment value associated with the text as a whole, which makes it difficult to track sentiment around any individual entity. With our toolkit, one could automatically extract “positive” and “negative” aspects within a given text, giving a clearer understanding of the sentiment surrounding its individual components.

Feel free to use the code, contribute to it, and come up with addtional applications; our system and models are available at https://github.com/yahoo/FEL.

[alfakilopapa]

This is a voice based cityscape navigation system. The front end has been written in javascript and html based on the boiler-plate example available from freeswitch software distribution. The backend is written in python integrated with navigation database APIs such as been written in javascript and html based on the boiler-plate example Google Places and GeoNames.org.

Source code: https://bitbucket.org/aasish/streetview_talkie